Outdoor physical activity, residential green spaces and the risk of dementia in the UK Biobank cohort

Dementia affects around 50 million people globally and is projected to reach 150 million by 2050, creating substantial personal and societal burden. Physical activity is known to benefit cognition and reduce dementia risk, but it is unclear how the setting of activity—particularly outdoors in natural environments—affects long-term brain health and dementia risk. Green exercise (physical activity in natural settings) may confer additional benefits through exposure to green environments, yet long-term evidence from large cohorts is limited. Prior experimental studies suggest short-term cognitive and neuroimaging benefits of outdoor activity, and observational studies link greater residential green space (GS) with lower risks of cognitive decline and dementia. This study aimed to quantify the association between outdoor physical activity and incident dementia among older adults, assess whether residential GS modifies this association, and explore neuroimaging correlates potentially underlying these relationships.

Experimental work demonstrates acute benefits of time outdoors for cognition compared with indoor activity, including improved performance after short outdoor walks and associations between time outdoors and greater gray-matter volume in prefrontal regions. A natural versus urban walk study reported hippocampal subfield volume increases after a one-hour nature walk. Systematic reviews note low-quality evidence for long-term benefits of outdoor activity in nature, underscoring a need for robust cohort studies. Large observational studies have linked higher residential GS with reduced cognitive decline and dementia risk, potentially via reduced cardiometabolic burden and slower vascular aging. GS may also facilitate and motivate outdoor activity, further benefiting vascular health and cognition. However, modification of the outdoor activity–dementia association by GS and long-term links to neuroimaging markers remain underexplored in large cohorts.

Design and population: Prospective cohort analysis within UK Biobank. From 502,198 enrollees (2006–2010), exclusions included dementia before baseline (n=258), invalid follow-up (n=39,170), and missing physical activity data (n=43,752), yielding 419,018. For dementia risk analyses, participants aged ≥60 years were considered; the analytical sample was 187,724. For neuroimaging analyses, all with brain MRI were included (n=36,854; imaging 2014–2021). Follow-up for incident dementia continued until December 31, 2022. The study adhered to STROBE guidelines and UK Biobank ethics approvals; informed consent was obtained. Exposures: Outdoor physical activity measured at baseline (2006–2010) and at neuroimaging visit (2014–2022) using the short-form IPAQ across leisure, domestic, occupational, and transport domains. Outdoor-related items included: walking for pleasure; light DIY (e.g., home improvement, gardening); heavy DIY (e.g., using heavy tools, weeding, lawn mowing, digging, carpentry); other exercise (e.g., swimming, cycling). Weekly metabolic equivalent of task (MET)-minutes were calculated for total outdoor activity and by item. Residential GS exposure was estimated as percent green area within a 300 m buffer around home address, using the 2005 Generalised Land Use Database (England) and the 2007 Land Cover Map (Wales, Scotland); area-weighted means were assigned per participant. Outcomes: Incident all-cause dementia and subtypes (Alzheimer’s disease, vascular dementia, frontotemporal dementia) were ascertained via linked primary care, hospital inpatient, death registers, and algorithmically defined first-occurrence data using ICD coding. Neuroimaging markers (2014–2021) included neurodegenerative markers—normalized total gray-matter volume; hippocampal volume (sum of left/right, weighted by total brain tissue volume)—and a cerebrovascular marker—total white-matter hyperintensity (WMH) volume (log-transformed). Imaging used Siemens Magnetom Skyra 3T with UK Biobank’s centralized processing pipeline. Covariates: Demographics (age, sex, education), smoking (current/former/never), cardiovascular health (diabetes, high cholesterol, hypertension), somatic health (Charlson Comorbidity Index), residential length, area-level socioeconomic status (Index of Multiple Deprivation), environmental exposures (air pollution 2005–2007; noise 2009). BMI was not included in Cox models due to violation of proportional hazards. Missing covariate data were handled via indicator categories. Statistical analysis: Baseline characteristics were summarized and compared by incident dementia status. Cox proportional hazards models (time-on-study as time scale) estimated associations between outdoor activity (quartiles and log-transformed continuous per 100 MET-min/week) and dementia outcomes. Model 1 adjusted for individual-level covariates; Model 2 additionally for IMD, air pollution, and noise. Effect modification by GS was tested via interaction term; where significant, stratified analyses contrasted top GS quartile (more accessible GS) versus remaining quartiles (less accessible GS) with three adjustment sets (Models 1–3). Associations between GS and activity (including specific items) were explored using deciles, density plots, and ordinal/multinomial logistic regression. For neuroimaging subsample (≥40 years at baseline), multivariable linear regressions estimated associations of outdoor activity and GS with hippocampal volume, total gray matter, and WMH, with progressive adjustments. Analyses were conducted in R 4.2.2 using survival, MASS, nnet, and stats packages. Sensitivity analyses included alternative natural environment metrics (blue space, domestic gardens, natural environment), 1000 m GS buffer, and excluding dementia cases in first 10 years to address reverse causality.

- Sample and follow-up: Among 187,724 participants aged 60–73 years at baseline (mean age 64.2; 52% female), 7,218 incident dementia cases occurred over a mean 13.01 years (SD 2.52) of follow-up. MRI subsample included 36,854 participants. - Primary associations: Higher outdoor activity was associated with lower all-cause dementia risk. Per log-transformed 100 MET-min/week, HR=0.93 (95% CI 0.91–0.95, p<0.001). Comparing quartiles, versus Q1: Q2 HR 0.94 (0.88–1.00), Q3 HR 0.88 (0.83–0.95), Q4 HR 0.84 (0.78–0.90). - Dementia subtypes: Vascular dementia showed the strongest association (continuous HR 0.88, 95% CI 0.84–0.92; quartiles vs Q1: Q2 0.83 [0.73–0.95], Q3 0.78 [0.68–0.89], Q4 0.72 [0.63–0.83]). Alzheimer’s disease showed a weaker association (continuous HR 0.96, 95% CI 0.93–1.00); quartile Q4 vs Q1 HR 0.93 (0.84–1.02). Frontotemporal dementia was not significant (continuous HR 0.90, 95% CI 0.78–1.04). - Effect modification by green space: Interaction between GS and outdoor activity for all-cause dementia was significant (p for interaction ≈0.045). In stratified analyses, the protective association of high outdoor activity (Q4 vs Q1) was stronger in areas with more accessible GS (top quartile): HR 0.73 (95% CI 0.63–0.85, p<0.001) versus less accessible GS: HR 0.86 (95% CI 0.79–0.93, p<0.001). For activity items, light DIY showed a stronger association in high GS areas (Q4 vs Q1 HR 0.77 [0.63–0.95], p=0.02) than in low GS (HR 0.93 [0.83–1.04], p=0.21). Walking was not protective in high GS but in low GS the highest walking quartile was associated with slightly increased risk (HR 1.15 [1.05–1.26], p=0.003). - Activity and GS relationship: Higher neighborhood GS was associated with higher engagement in outdoor activities overall and for walking, light DIY, and heavy DIY; regression analyses supported these trends. - Neuroimaging: Greater outdoor activity was associated with larger hippocampal and total gray-matter volumes and fewer WMH. Continuous outdoor activity was positively associated with hippocampal volume (β ~7.78 mm³, 95% CI 0.57–14.99 in basic adjustment; attenuated with fuller adjustment) and with total gray matter (β ~696–726 mm³ across models; p<0.001), and inversely with WMH (β ~−0.03, p<0.001). Stratified by GS, associations with neurodegenerative markers, particularly hippocampal volume, were more pronounced in high-GS areas, while WMH associations did not differ by GS. - Sensitivity analyses: Results were robust when considering blue space, domestic garden percentage, broader natural environment, a 1000 m GS buffer, and after excluding dementia cases in the first 10 years of follow-up.

The study demonstrates that higher engagement in outdoor physical activity among older adults is associated with a reduced risk of incident dementia, particularly vascular dementia, and to a lesser extent Alzheimer’s disease. Residential green space appears to amplify this protective association, suggesting that environmental context can enhance the brain-health benefits of physical activity. Findings link outdoor activity to favorable neuroimaging profiles—larger hippocampal and gray-matter volumes and fewer WMH—supporting both neurodegenerative and vascular pathways as plausible mechanisms. The stronger association with hippocampal volume in high-GS neighborhoods points to synergistic effects of nature exposure and activity on neural plasticity and stress regulation, consistent with prior experimental evidence on nature’s effects on brain function and rumination. The observed positive relationship between GS availability and outdoor activity engagement indicates that greener neighborhoods may facilitate and motivate physical activity, aligning with conceptual models and public health frameworks (e.g., WHO Age-Friendly Cities). Collectively, these findings underscore the importance of integrating accessible, well-designed green spaces into urban planning to promote healthy behaviors and reduce dementia risk in aging populations.

Outdoor physical activity is associated with lower long-term risk of dementia in older adults, with the strongest effects observed for vascular dementia. Accessible residential green space enhances these benefits and is associated with better preservation of neuroimaging markers, particularly hippocampal and total gray-matter volumes. The results support strategies that combine promotion of outdoor activity with urban designs that increase equitable access to green and natural environments. Future research should employ objective, longitudinal measures of outdoor activity (e.g., wearables), refine classification of indoor versus outdoor activity, leverage diverse biomarkers to elucidate neurodegenerative and vascular mechanisms, test different spatial buffers and activity contexts, and replicate findings in more diverse populations and settings.

- Volunteer and selection bias in UK Biobank may limit generalizability; dementia incidence was somewhat lower than population-based estimates, and the cohort overrepresents individuals with white ethnicity and higher education. - Sex differences in dementia risk in this cohort differ from typical patterns, potentially reflecting differential health behaviors and conditions. - Dementia ascertainment relied on linked routine healthcare data and algorithms; while positive predictive value is high, misclassification of subtypes remains possible, and some subtypes (e.g., Lewy body dementia) are not well-captured by ICD-10 codes. - Limited statistical power for frontotemporal dementia analyses. - Outdoor activity was self-reported (IPAQ), introducing potential recall and reporting biases; “other exercise” may include indoor activities (e.g., indoor swimming), complicating outdoor-specific inference. - Lack of time-varying residential history and potential exposure misclassification due to relocation; GS derived from 2005 land-use data, and air/noise estimates from 2005–2009 may not reflect changes through 2022. - BMI was excluded from Cox models due to proportional hazards violation; with nearly 70% overweight/obesity, residual confounding and selection bias are possible. - Potential unmeasured confounding and measurement error remain despite extensive covariate adjustment and sensitivity analyses.